[katzinsky]

That's surprisingly low considering it was probably trained on many of the papers it's supposed to be replicating.